[Fix] StructToData: allow empty values through for Optional+Computed fields

[diraol]

Problem

StructToData in common/reflect_resource.go has a gate that skips writing empty/nil API values to state for Optional fields:

// Before
if fieldSchema.Optional && isValueNilOrEmpty(valueField, fieldPath) {
    return nil
}

This gate is correct for Optional-only fields: if the API returns a server default the user never configured, we should not pollute state with it.

However, the same gate was also firing for Optional+Computed fields — ones where the server is explicitly allowed to set the value. When the API returned an empty map or nil for such a field (e.g. a cluster policy clearing spark_conf), the value was silently dropped, never reaching Terraform state. This caused two concrete problems:

1. Perpetual (known after apply) diffs — state never reflected the empty server value, so Terraform kept re-planning it as unknown.
2. lifecycle { ignore_changes } not working — because the field's current server value was never written to state, the provider could not track it, and on the next apply it would silently revert the externally-set value back to whatever was last in state.

Fix

Add !fieldSchema.Computed to the gate condition:

// After
if fieldSchema.Optional && !fieldSchema.Computed && isValueNilOrEmpty(valueField, fieldPath) {
    return nil
}

This preserves existing behaviour for Optional-only fields (server defaults are still filtered) while allowing Optional+Computed fields to write empty/nil values through to state.

Tests

• TestStructToData_OptionalComputedEmptyMapWrittenToState — regression test: empty map for an Optional+Computed field must reach state.
• TestStructToData_OptionalOnlyEmptyMapNotWrittenToState — guard test: empty map for an Optional-only field is still filtered out.

Notes

This is the prerequisite central fix. Per-resource PRs (marking policy-settable fields as Optional+Computed) will follow and depend on this change.

Related: databricks#1238